792032 - Display glitches in Blobular SVG demo with hardware acceleration enabled

Status: RESOLVED WORKSFORME

(Core :: Graphics, defect)

Description

[Loic]

Attachment: screenshot FF17 (1st regression) and FF18 (2nd regression)

Blobular SVG demo: http://www.themaninblue.com/experiment/Blobular/
Drag the circle from the center and move the mouse to detach a 2nd circle.
During the animation, there are display glitches that disappear when the user stops moving the mouse.

The issue is not visible with HWA disabled.

I found 2 regression ranges, see my screenshot to have a better idea.
After the 1st regression, the demo is slightly broken. The 2nd regression increases a lot the issue.

Regression #1:
m-c
good=2012-08-02
bad=2012-08-03
http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=588424024294&tochange=89dcadd42ec4

Suspected bug (need bisecting maybe)
Nicholas Cameron — Bug 779650; handle Cairo DrawTargets in CanvasLayerD3D10; r=bas

Regression #2:
m-c
good=2012-09-16
bad=2012-09-17
http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=f7c89de3ab43&tochange=3f0587ce1774

Suspected bug:
Matt Woodrow — Bug 788044 - Make transformed frames the reference frame for their display list tree. r=roc

Comment 1

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Yours, Nick.

Comment 2

[Nick Cameron [:nrc]]

I'll look into it.

My suspicion is that 779650 is not to blame since that only touches canvas layer code (and not very much at that) and this looks like a pure SVG demo. If I were to hazard a guess, I would suspect Bug 776054, since that looks like it could affect a lot of SVG stuff and is kind of related to 788044.

Comment 3

[Jonathan Watt [:jwatt]]

Flipping svg.display-lists.painting.enabled to false does indeed seem to fix this. Assigning to myself since I assume you'd rather I take this, Nick, but if you want to dig into it feel free to take it back. :)

Comment 4

[Loic]

So it looks like my guess for the 1st suspected bug is false, bug 776054 is surely the culprit. Is a better bisection necessary?

Comment 5

[Nick Cameron [:nrc]]

(In reply to Jonathan Watt [:jwatt] from comment #3)
> Flipping svg.display-lists.painting.enabled to false does indeed seem to fix
> this. Assigning to myself since I assume you'd rather I take this, Nick, but
> if you want to dig into it feel free to take it back. :)

Thanks Jonathan! I'm not particular attached to it. Feel free to send it back my way if it is not an SVG thing.

Comment 6

[Alex Keybl [:akeybl]]

Tracking as SVG is obviously being used more and more in web apps, and we don't know where else there may be fallout.

Comment 7

[Alex Keybl [:akeybl]]

jwatt - any updates on this investigation?

Comment 8

[Jonathan Watt [:jwatt]]

With DLBI we invalidate perfectly (and the smallest possible area, which is great). I'm not sure if it's worth fixing this for 17 given the misrendering is only transient and I haven't seen the issue in any other SVG yet. At any rate, I'm not going to have a fix for this before the imminent uplift.

Comment 9

[Alex Keybl [:akeybl]]

Given comment 8, untracking.

Comment 10

[Loic]

It seems to be fixed in FF17+, I'm not able to reproduce the glitches observed in my screenshot (cf https://bugzilla.mozilla.org/attachment.cgi?id=662145).
Can someone confirm too?

The only remaining issue is bug 792058.

Comment 11

[Jonathan Watt [:jwatt]]

Strange. I can confirm that this isn't an issue in Firefox 17.0.1 or 17.0.1 ESR, or in Firefox 17 beta 1 through beta 6 (that's all of them). In Nightly builds I can reproduce it right up until DLBI (bug 539356) landed. Since DLBI is only in 18, not 17, I don't know what fixed this for 17. Since we don't know what fixed it, closing as worksforme.